Assessing among-locus variation in the inference of seed plant phylogeny

Large multilocus analyses can greatly reduce sampling error in phylogenetic estimates and help resolve difficult phylogenetic questions. Yet conventional multilocus analyses may be confounded by variation in the phylogenetic signal or processes of evolution among loci. We used nonparametric bootstrapping methods to examine locus-specific variation within a 12-locus seed plant data set and to examine the effects of this variation on estimates of seed plant phylogeny. The observed maximum likelihood and maximum parsimony bootstrap support from phylogenetic analyses of sites within single loci often notably differs from the bootstrap support obtained by sampling an equal number of sites from the concatenated 12-locus data set. This indicates heterogeneity among loci in the phylogenetic inference, and the differences among loci are not explained by the distribution of fast and slowly evolving sites. Bootstrap analyses that resample loci with replacement, rather than sampling individual sites with replacement, reveal extensive sampling variance among loci. The results suggest that seed plant phylogenetic analyses may not be robust to sampling error when only 12 loci are used and indicate a need for further investigation into the causes of the locus-specific variation.